基于U-net卷积神经网络的大转角ISAR成像方法

针对ISAR成像在大转角条件下产生严重的越距离单元徙动从而使得ISAR图像散焦的问题,提出一种基于U-net卷积神经网络的大转角ISAR成像方法。首先利用快速傅里叶变换对大转角条件下的回波数据进行预处理,得到散焦的ISAR复值图像作为训练样本,其次,根据ISAR成像特点对U net网络结构进行了改进,训练后得到具有良好聚焦能力的成像网络。仿真实验表明：与传统大转角ISAR成像方法相比,所提方法将ISAR图像的峰值旁瓣比降至-18 dB以下,具有更小的图像熵和最小均方误差,成像时间缩减至0.28 s左右,在低信噪比条件下仍可以实现ISAR图像的快速、准确重建。

Wide Angle ISAR Imaging Based on U-net Convolutional Neural Network

In wide angle inverse synthetic aperture radar (ISAR) imaging, serious migration through range cells (MTRC) will lead to the defocus of ISAR image. A wide angle ISAR imaging method based on U-net convolutional neural network (U-net CNN) is proposed, Firstly, the echo data is preprocessed by fast Fourier transform to obtain a defocused ISAR complex value image as the training samples; Secondly, according to ISAR imaging characteristics, the u-net structure is improved, and an imaging network with good focusing ability is obtained after training. Simulation results show that compared with traditional wide angle ISAR imaging methods, the proposed method reduces the peak sidelobe ratio (PSLR) of ISAR image to less than 18 dB, has smaller image entropy and minimum mean square error (NMSE), and the imaging time is reduced to about 0.28 seconds. Under the condition of low signal to noise ratio (SNR), the proposed method can still achieve fast and accurate reconstruction of ISAR image.